Computers and Geotechnics 48 (2013) 117126

Contents lists available at SciVerse ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Finite element simulation of an embankment on soft clay Case study

Jinchun Chai a,, Yutaka Igaya b,1, Takenori Hino c,2, John Carter d,3
a
Department of Civil Engineering and Architecture, Graduate School of Science and Engineering, Saga University, Japan
b
Prefectural Planning Head Ofce, Saga Prefectural Government, Japan
c
Institute of Lowland and Marine Research, Saga University, Japan
d
Faculty of Engineering and Built Environment, The University of Newcastle, NSW 2308, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Numerical simulations and eld measurements of an embankment constructed on a deposit of soft Ariake
Received 14 February 2012 clay in Saga, Japan are compared and discussed. The simulations were made both before (Class-A) and
Received in revised form 8 October 2012 after (Class-C) the eld data became known. It is shown that the Class-A prediction resulted in poor sim-
Accepted 8 October 2012
ulations of the measured settlementtime curves, mainly due to over-estimation of the magnitude of the
yield stresses of the subsoils (i.e., the sizes of the yielding loci) and under-estimation of the compressibil-
ity, hydraulic conductivity and the slope (M) of the critical state line. It is demonstrated that: (a) appro-
Keywords:
priate site investigation, soil testing and interpretation of the test results are essential for accurate
Soft ground
Finite element analysis
prediction of the behaviour of an earth structure constructed on soft clayey deposits; (b) when using a
Embankment soil model developed within the framework of Critical State Soil Mechanics to make such predictions,
Case history M value should be directly determined from tests with an appropriate effective stress path; and (c) yield
stresses of soft soil layers can be calibrated by comparing the predicted undrained shear strengths (Su)
with measured values, provided the effect of strain rate and/or strain softening on the value of Su is prop-
erly considered. The results of this analysis indicate that Bjerrums strain rate correction factor can be
adopted as a rst approximation of the correction factor applied to eld or laboratory measured values
of Su.
2012 Elsevier Ltd. All rights reserved.

1. Introduction
Comparing the predicted and eld-measured behaviour of
embankments constructed on soft clayey ground often provides a
good check on the suitability of the constitutive model adopted
for the clay soil, as well as the capacities of the numerical proce-
dures used to make predictions of the embankment behaviour,
providing of course that appropriate values have been adopted
for the model parameters. The latter is a function of the quality
of sampling, sample preparation, testing and overall characterisa-
tion of the soils at the site. There is a relatively rich literature in
this particular eld of geotechnical engineering (e.g.,
[30,22,4,6,9,12,15,20,16,21,18]).
The Modied Cam Clay (MCC) constitutive model [23] is one of
the most widely used elastoplastic models for soft clayey soils be-
Corresponding author. Tel.: +81 952 28 8580; fax: +81 952 28 8190.
E-mail addresses: chai@cc.saga-u.ac.jp (J. Chai), igaya-yutaka@pref.saga.lg.jp (Y.
Igaya), hino@ilt.saga-u.ac.jp (T. Hino), John.Carter@newcastle.edu.au (J. Carter).
1
Tel.: +81 952 66 0912; fax: +81 952 66 0956.
2
Tel./fax: +81 952 28 8612.
3
Tel.: +61 2 4921 6025.
cause of its simplicity and its ability to predict yielding, strain soft-
en as well as failure in soft clayey soils. However, MCC is an
isotropic yielding model and it does not consider viscous behaviour
such as creep of clayey soils. Elasto-viscoplastic (EVP) models (e.g.,
[31]) and anisotropic yielding elastoplastic models (e.g., [11]) have
been developed to account for these complicating effects. How-
ever, there are still differing opinions as to which model provides
the best prediction of the eld response of clayey soils under
embankment loading (e.g. [18,20]). In principle, a more sophisti-
cated soil model should be able to represent better the mechanical
behaviour of natural soft clayey soils, but the drawback in adopting
them is that more sophisticated models require specication of
more soil parameters, and in many cases in engineering practice
there are insufcient test data to reliably dene those parameters.
Further, for some sophisticated soil models, some of the soil
parameters can only be calibrated by tting the model predictions
to the test results, rather than being determined directly from
those test results. On the other hand, the parameters of relatively
simple soil models, such as the MCC model, which is acknowledged
for being capable of capturing the most important mechanical fea-
tures of soft clayey soil, can be easily and reliably dened. It is gen-
erally acknowledged that the use of some simple models can result
in acceptable predictions of soil behaviour, at least from a practical

0266-352X/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compgeo.2012.10.006
118 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126

Fig. 2. Borehole locations in and around Test Site.

Fig. 1. Location of Ariake Sea Coastal Road and test embankment.

Grain size t
composition wp wn wL
(%) (%) (kN/m3 ) e0
perspective. Whichever the case, obtaining accurate Class-A pre- Soil 0 50 0 100 10 15 0 1 2 3 4
0 Surface
dictions [17] of the behaviour of embankments on soft clayey
layer
deposits still remains a difcult task. C M

As shown in Fig. 1, in Saga, Japan, a highway around the Ariake

BH-6
Sea has been planned and was under construction at the time of
writing. For its entire length this highway is located over deposits Soft
of soft Ariake clay. In order to verify the correctness of the assumed 5 silty
clay
Depth (m)

design strength and deformation parameters of the soft deposit, as

estimated from laboratory tests, a test embankment was built on
the natural deposit and its performance was monitored for more
than 3 years in terms of settlements, lateral displacements and ex-
cess pore water pressures. For this test embankment an internal
Class-A prediction was made and documented before construction.
After the eld-measured data became known, these Class-A predic-
tions were compared with the measurements and there were con-
siderable discrepancies. The same case was then re-analyzed
(Class-C prediction) using additional site investigation results and
after applying a correction to the measurements of the undrained
shear strength (Su) of the subsoils. For all the analyses reported
here, the soft Ariake clay was modelled by the MCC model.
In this paper the site conditions, history of the embankment
construction and the measured data are reported and compared to-
gether with the results of the Class-A and Class-C predictions. The
insights gained into predicting the behaviour of an embankment
on soft ground are discussed.
2. Soil prole and embankment construction
In total, three (3) test embankments were constructed at the
location indicated in Fig. 1, one on natural soft ground and two
on the same type of soft ground after it had been improved by
the installation of soilcement columns formed by deep mixing
[13]. The test embankment considered in this study is the one con-
structed on natural ground. As shown in Fig. 2, in and around the
test site, six (6) boreholes (BH) were drilled in order to investigate
the soil properties required for design of the road embankment. At
the locations of BH-1, -3, -5 and -6, undisturbed soil samples were
taken using a Japanese thin-wall sampler and laboratory index
tests and consolidation and unconned compression tests were
conducted on these samples.
BH-1 to BH-5 were drilled and the corresponding laboratory
tests were conducted before the test embankment was con-
structed, and the resulting data were available at the time the
Class-A prediction was made. BH-6 was in the test site and was
bored just before the commencement of embankment construc-
tion. The test data from this borehole were not available when
Silty
clay
10
Organic
clay
S G
Clayey
sand
C: clay BH-6
M: silt BH-1, 3
S: sand
G: gravel &5
15
Fig. 3. Soil prole and some physical and mechanical properties at the Test Site.
the Class-A predictions were made. The soil prole and some phys-
ical properties obtained from BH-1, 3, 5 and 6 are summarized in
Fig. 3. In this gure, Wp, WL and Wn are plastic limit, liquid limit
and natural water contents respectively, ct is the unit weight,
and e0 is the initial void ratio. Values of Wp and WL are only for
the samples from BH-6.
It is noted that at this site there exists a surface layer about
1.5 m thick underlain by a thick soft silty clay layer (the Ariake
clay) which is about 8.0 m thick. Below it is an organic clayey soil
layer about 0.3 m thick, underlain by alternating clayey sand and
sandy clay layers. The natural water content of the soft silty clay
was generally more than 100% and larger than the corresponding
liquid limit. The groundwater level was about 1.0 m below the
ground surface.
For the undisturbed samples from BH-6, consolidated un-
drained triaxial compression tests with excess pore water pressure
measurement were also conducted. The effective stress paths in a
p0 q plot (p0 is effective mean stress and q is deviator stress) of
the triaxial test are given in Fig. 4. In the gure, r01 and r03 are effec-
tive stresses in the vertical and horizontal directions respectively.
The e- logr0v ) curves (e is void ratio and r0v is vertical consolidation
stress) from odometer test results for the samples from BH-6 are
given in Appendix A.
For the test embankment constructed on natural ground the
critical height was estimated to be about 3.0 m, and so it was
decided that an embankment should be constructed with an initial
 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126 119

Fig. 4. Effective stress paths.

ll thickness of 2.5 m. Once this was achieved, and depending on 3. Predictions and comparison with eld measurements
the results of the ground monitoring, a decision was then to be
made whether to increase the ll thickness to 3.0 m. In practice 3.1. Finite element modelling
the embankment was only constructed to a total ll thickness of
2.5 m. After the ll reached this thickness the monitored results Predictions were made using plane strain nite element analy-
indicated that the rate of lateral displacement was increasing ses (FEA). The nite element mesh and the boundary conditions
quickly and after stopping the placement of ll the lateral move- adopted are illustrated in Fig. 6. Due to symmetry, only half the
ment continued with the rate for about a week. The supervisor in embankment was modelled, and the modelled area had an overall
charge of this project decided to cease further lling altogether, be- horizontal width of 60.0 m and a vertical thickness of 22.5 m. At
cause building the embankment to failure was not an option. The the left and the right boundaries the horizontal displacement
base dimensions (length  width) of the embankment were was xed but vertical displacement was allowed. At the bottom
46.8 m  21.8 m and its side slopes were 1:1.8 (V:H), which re- boundary both the horizontal and vertical displacements were
sulted in plan dimensions of the top of the embankment of xed. Both the ground surface and the bottom boundaries (sand
37.8 m  13.8 m. Decomposed granite was used as ll material layer) were considered completely permeable, i.e., they were
and the average lling rate was about 0.05 m/day. The average to- drainage boundaries, and both the left and right boundaries were
tal unit weight of the embankment ll as placed was about considered to be impermeable. Eight-noded quadrilateral elements
19.0 kN/m3. A cross-section of the embankment and some of the with excess pore water pressure degrees of freedom at only the
key instrumentation points for measuring settlements, lateral dis- four (4) vertex nodes were used to represent the foundation soil.
placements and pore water pressures are shown in Fig. 5. The eld- Eight-noded quadrilateral elements without the excess pore water
measured data presented in the following sections were all sourced pressure degrees of freedom were used to represent the embank-
from the [1]. ment. The adequacy of the adopted mesh was checked by also
using a much ner mesh, by almost quadrupling the number of
13.8 m elements. By comparison of solutions it was found that the mesh
originally adopted is capable of providing accurate and converged
2.5 m S0
predictions of the nodal deformations.
0.0 m In the numerical analyses, the soft clayey layers were repre-
Surface layer P1 (1.4 m)
1.5 sented by the Modied Cam Clay (MCC) stressstrain model [23],
S1 (2.8 m)
Soft silty clay P2 (5.0 m) and the sand layers and the embankment ll material were simu-
S2 (7.1 m) Inclino-
P3 (8.3 m) meter lated by an elastoplastic stressstrain model which obeys the
9.5 Organic clayey soil S3 (9.5 m)
Mohr-Coulomb failure criterion.
9.8
P4 (10.0 m) S4 (12.0 m) The predictions were conducted using a fully coupled nite
Clayey sand
P5 (13.0m) Settlement plate element consolidation analysis and the program used for the
14.5
16.0 Stiff sandy clay Pore pressure gauge numerical calculations was CRISP-AIT [7], which is based on the
Clayey sand original CRISP program [5]. Large deformations were considered
20.5
22.4 Volcanic ash layer
approximately by updating the nodal coordinates at the end of
each incremental step. Further, to ensure that the true stress
S5 (24.0 m) To 24 m strain law is followed closely in the numerical model, the New-
tonRaphson method with an explicit sub-stepping technique
Fig. 5. Cross-section of the embankment and key instrumentation points.
120 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126

Fig. 6. Mesh and boundary conditions.

that includes error control [26] has been incorporated into the -
nite element program.
3.2. Class-A predictions and comparisons with measured data
3.2.1. Model parameters
The model parameters adopted for the Class-A prediction were
estimated from the laboratory test results using the undisturbed
samples obtained from BH-1, -3, and -5. The location of these bore-
holes is referred to in the following as the Nearby Area. The param-
eter values deduced from the laboratory testing are listed in Table 1.
In this table the values of Poissons ratio, m, were assumed to be 0.3.
Values of j were assumed as 0.1k. At the time of making the Class-A
prediction, there were no test results that allowed an independent
measure of the friction angle (/0 ), and so the slope of the critical
state line in p0 q space, M, was calculated by assuming a /0 value
of 30 for triaxial compression conditions (based on local experi-
ence). The values of Youngs modulus (E) of the sandy layers were
estimated from standard penetration test N-values as
E = 2500N (kPa) [14]. The values of kv were estimated as twice the
values deduced from the laboratory incremental loading consolida-
tion tests, and the values of horizontal hydraulic conductivity, kh,
were set as 1.5 times the corresponding value of kv, based on previ-
ous experience with this soil [8]. The values of kv and kh listed in Ta-
ble 1 are initial values and during consolidation they were allowed
to vary with void ratio according to the following equation [29]:
k k0  10e0 e=Ck
where k0 is the initial hydraulic conductivity, e0, the initial void ra-
tio, k, the current hydraulic conductivity, e, the current void ratio,
and Ck is a constant. Tavenas et al. [28] suggested that generally
Ck = (0.40.5)e0 and in this study Ck = 0.4e0 was assumed.
The overconsolidation ratio (OCR) of the soft soil is a key param-
eter controlling the amount of deformation of the deposit under
the embankment loading. It is linked with the size of yield locus
(p0y ) and therefore the undrained shear strength (Su), as predicted
by the MCC model. According to this model and for undrained tri-
axial compression stress paths, Su value predicted by MCC can be
expressed by the following equation:
!K
p0 M 2 g2
Su 1K
M OCRK 2
2 M2
where K = 1  j/k, and g = q/p0 . Determining the value of the pre-
consolidation stress (pc) and therefore the value of OCR from the
incremental loading laboratory consolidation test results may de-
pend on the judgment of the person interpreting the test results.
To avoid completely subjective judgment, the approach adopted
was to adjust the value of pc by comparing the measured value of
Su with the value predicted by Eq. (2). The measured value of Su
was assumed to be Su = qu/2, where qu is the measured unconned
compression strength of the undisturbed soil sample. In order to
predict the value of Su from the MCC model, Mayne and Kulhawys
[19] equation was used to calculate the coefcient of earth pressure
at-rest (K0) and therefore the initial effective stress in the ground in
the horizontal direction. This value of K0 was computed by assum-
1 ing an internal friction angle of the subsoil of 30.
Measured data from BH-1, -3, and -5 (open circles) and the pre-
dicted values (dashed lines) are shown in Figs. 7a and 8a for Su (la-
belled Su-A&C1) and OCR (labelled OCR-A), respectively. The

Table 1
Model parameters.

Depth (m) Soil strata SPT N E (kPa) m j k M e0 ct (kN/m3) kv kh

(104 m/day)
0.01.5 Surface soil 0.30 0.025 0.25 1.2 1.50 16.0 6.0 9.1
1.54.0 Soft silty clay 0.30 0.065 0.65 1.2 3.14 13.7 5.1 7.7
4.06.0 0.30 0.059 0.59 1.2 2.89 13.9 5.4 8.1
6.08.0 0.30 0.060 0.60 1.2 2.67 14.1 5.4 8.1
8.010.0 0.30 0.071 0.71 1.2 2.55 14.3 4.6 6.9
10.012.0 Sandy clay 0.30 0.008 0.08 1.2 1.10 18.0 17.5 26.3
Depth (m) Soil strata SPT N E (kPa) m / () C (kPa) M e0 ct (kN/m3) kv kh
12.015.0 Clayey sand 8 20,000 0.25 35 20 0.80 18.0 2500 2500
15.020.0 Clayey sand 15 37,500 0.25 35 20 0.70 19.0 2500 2500
Embankment 2 5000 0.30 35 20 19.0

Note: E = Youngs modulus; m = Poissons ratio; k = slope of consolidation line in e-ln p0 plot (e is voids ratio and p0 is effective mean stress); j = slope of rebound line in e-ln p0
plot; M = strength parameter for Cam-clay model, stress ratio at failure, qf/p0 (q is deviator stress); e0 = initial void ratio; ct = unit weight; kh and kv = horizontal and vertical
hydraulic conductivity, c = cohesion, and /0 = friction angle of soil.
 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126 121

Su & Su' (kPa) Su & Su' (kPa)

0 10 20 30 40 0 10 20 30 40
0 0
Nearby Area Test Site

2 2

S u-A&C1
4 4

Depth, D (m)

Depth, D (m)
Su-A&C1

6 6

8 8 Su-C2
Su-C2

10 10
Measured (Su) Measured (S u)
Corrected (Su') Corrected (Su')
12 12
(a) Nearby area (b) Test site
Fig. 7. Variation of Su.

OCR OCR
0 2 4 6 8 0 2 4 6 8
0 0

Nearby Area 2
Test Site
2

4 4
Depth, D (m)
Depth, D (m)

6 6

8 8

Measured Measured
10 OCR-A 10 OCR-A
OCR-C1 OCR-C1
OCR-C2 OCR-C2
12 12
(a) Nearby area (b) Test site
Fig. 8. Variation of OCR.

predicted values of Su are slightly lower than the average measured
values, but the predicted values of OCR form almost an upper
bound on the measured data. Initially these discrepancies were
judged to be due to unavoidable sample disturbance, i.e., the mea-
sured OCR values may have been less than the actual eld values.
The values of Su-A&C1 and OCR-A shown in Figs. 7a and 8a
were adopted in the Class-A prediction of the embankment
response.
3.2.2. Comparison of predictions and eld measurements
A comparison of the settlementtime curves is given in Fig. 9ac
for settlement gauges S0, S1 and S3, respectively (see Fig. 5 for the
locations). There were some problems at settlement gauge S2 and
the data were considered unreliable and there were only very small
settlements measured at the location of S4 (about 18 mm), so that
comparisons for these two locations have been omitted. It can be
seen that as for the settlement at the ground surface (S0), the
Class-A prediction is at most about 50% of the measured data, i.e.,
a poor Class-A prediction has been obtained for these settlements.
The simulated settlement at a depth of 9.5 m (Fig. 9c) reduced
slightly during the consolidation process. Although it is not very
obvious, the measured data also show the same tendency. The rea-
son is probably related to the likelihood that the plastic strain was
concentrated largely in the very soft layer, and squeezing the soil
sideways meant that there might be slight stress decrease deeper
down. It is worth mentioning that from the settlement vs. log time
plot it is assumed that primary consolidation nished at about
3 years after the beginning of the construction.
The lateral displacement proles at the end of the embankment
construction and 2 years from the beginning of the construction
are compared in Fig. 10a and b, respectively. Although there are
discrepancies, compared with the settlement predictions it could
be concluded that the Class-A prediction has provided reasonable
estimates of the lateral displacements.
Comparisons of the measured and predicted excess pore water
pressures (u) at the locations P1 to P4 (see Fig. 5 for the locations)
are shown in Fig. 11ad, respectively. Very small values of pre-
dicted and measured excess pore water were relevant at location
P5 and so a comparison is not presented. It is clear that the
Class-A prediction method yielded good estimates of excess pore
water pressure (u) at locations P1 and P2, but under and over-pre-
dicted values of u at locations P3 and P4, respectively. Generally, the
Class-A prediction method resulted in acceptable simulations of
these excess pore water pressures.
122 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126

0 Based on the results of BH-6, values of the model parameters, k,

Measured j, e0, ct, kh and kv were redened for the soil layers from 1.5 m to
-0.2 Class-A
Class-C1
12.0 m, as listed in Table 2. Other modications are as follows:
Settlement, S (m)

-0.4 Class-C2
(1) Soil model: Based on the site investigation results of the
-0.6
Nearby Area, the soil layer from 10.0 m to 12.0 m depth
-0.8 was classied as sandy clay, but the results from the Test
Site indicate it is a clayey sand layer with a depth range of
-1
9.812.0 m. For this layer, the constitutive model was chan-
-1.2 ged from MCC to an elastoplastic stressstrain model that
S0 (Surface, center) obeys the Mohr-Coulomb failure criterion and the depths
-1.4
0 200 400 600 800 1000 1200 were changed accordingly.
Elapsed time, t (day) (2) M values: From the triaxial compression test results obtained
(a) Surface for four different depths at the Test Site (Fig. 4), the slope of a
straight line (sin(/0 )), drawn from the origin to the peak
0 point of each individual effective stress path, reduces as
-0.2 S1 (-2,8 m, center) the initial consolidation stress increases. Because MCC is a
no-cohesion model, an envelope of best t passing through
Settlement, S (m)

-0.4 the origin was determined by considering the representative

-0.6 stress levels in the deposit under the embankment loading.
The redened values of M (corresponding to a /0 value of
-0.8 about 39.2 under triaxial compression) are listed in Table 2.
-1 Measured
Class-A The sizes of the yield loci were subsequently redened in order
-1.2 Class-C1
Class-C2 to predict the target values of Su, as determined from the uncon-
-1.4 ned compression tests. However, the values of K0 used to calcu-
0 200 400 600 800 1000 1200
late the initial horizontal effective stresses in the deposit were
Elapsed time, t (day)
not changed, even though Mayne and Kulhawys [19] equation
(b) 2.8 m below the surface (S1) would suggest that a higher internal friction angle (/0 ) should re-
sult in a slightly smaller value of K0. Furthermore, it is noted that
Measured the results of constant rate of strain consolidation tests conducted
0.02 Class-A in the laboratory using undisturbed Ariake clay samples, cut verti-
Class-C1
Class-C2 cally and horizontally with respect to the in situ condition, indicate
Settlement, S (m)

0 that the yield stress in the horizontal direction is about 0.51.0

(average of about 0.7) times the value in the vertical direction
-0.02 [10]. Thus, for simplicity, it was assumed that values of K0 used
to dene the initial horizontal effective stress should be left at
-0.04 approximately 0.50.6 in the Class-C prediction.

S3 (-9.5 m) (3) Poissons ratio: Although the Class-A prediction resulted in a

-0.06
0 200 400 600 800
Elapsed time, t (day)
(c) 9.5 m below the surface (S3)
Fig. 9. Comparison of settlementtime curves.
3.3. Class-C predictions and comparisons with measured data
3.3.1. Analysis with soil parameters from BH-6
After the results of tests on soil samples recovered from bore-
hole BH-6 (whose location will be referred to as the Test Site) be-
came known, the correctness and appropriateness of the values
assumed for the model input parameters were checked. Compari-
sons of the adopted and measured values of Su and OCR are given
in Figs. 7b and 8b, respectively. Although the data are scattered,
the adopted values could be considered to be reasonable. However,
for the values of k there are considerable differences, as shown in
Fig. 12. For soil layers from the ground surface to about 9 m depth,
the values of k from BH-6 are generally much larger than those ob-
tained from samples of the Nearby Area. The corresponding initial
voids ratio, unit weight and hydraulic conductivity are also differ-
ent. It is notable that this site was tidal land about 300 years ago
[25]. Therefore, the very soft, more compressible spots may reect
the locations of the channels of old creeks or rivers.
1000 1200 good prediction of the lateral displacement prole, it was
considered that the lateral displacements might now be
over-predicted if the larger values of the compression
parameters, as listed in Table 2, were adopted in the revised
analysis. In the Class-C prediction, a value of Poissons ratio
of 0.15 was adopted for all soil layers (except the embank-
ment ll material).
(4) Size of the initial yield loci: Changing the value of M will obvi-
ously change the shape of the yield function. As illustrated in
Fig. 13, in order to predict the same target value of Su corre-
sponding to a given initial stress state, the size of the yield
locus (p0y ) for M = 1.6 is smaller than that corresponding to
M = 1.2. Also, with the assumed initial stress state and for
undrained triaxial compression, the effective stress path
for the case where M = 1.2 approaches the critical state line
(CSL) from the dry side, while for M = 1.6 it approaches CSL
from the wet side. Keeping the values labelled Su-A&C1
in Fig. 7 unchanged, the predicted OCR values were changed
to those labelled OCR-C1 in Fig. 8.
An analysis using the parameters in Table 2 was subsequently
conducted. The results of this analysis are given in Figs. 911,
where they are labelled as Class-C1 predictions. It can be seen that
there is considerable improvement in the settlement predictions,
but generally these predictions are still smaller than the measured
data (Fig. 9). Fig. 11a and b reveal an interesting point about the
 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126 123

Lateral displacement, (m) Lateral displacement, (m)

0 0.05 0.1 0.15 0 0.05 0.1 0.15
0 0

5 5

Depth, D (m)

Depth, D (m)
10 10

t = 50 days
(end of construction) t = 730 days
15 15 Measured
Measured
Class-A Class-A
Class-C1 Class-C1
Class-C2 Class-C2
20 20
(a) At the end of construction (b) At 2 years elapsed time
Fig. 10. Comparison of lateral displacement proles.

50 50
P1 (-1.4 m) Excess pore pressure, u (kPa)
Excess pore pressure, u (kPa)

P2 (-5.0 m)
40
Measured Measured
40
Class-A Class-A
Class-C1 Class-C1
30 Class-C2 30 Class-C2

20 20

10 10

0 0
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Elapsed time, t (day) Elapsed time, t (day)
(a) At 1.4 m below the surface (P1) (b) At 5.0 m below the surface (P2)

50 50
P3 (-8.3 m)
Excess pore pressure, u (kPa)
Excess pore pressure, u (kPa)

P4 (-10.0 m)
40 Measured 40
Class-A Measured
Class-C1 Class-A
30 Class-C2 30 Class-C1
Class-C2
20 20

10 10

0 0
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Elapsed time, t (kPa) Elapsed time, t (day)
(c) At 8.3 m below the surface (P3) (d) At 10.0 m below the surface (P4)

Fig. 11. Comparison of excess pore water pressuretime curves.

variation of u at location P1 (1.4 m from the ground surface) and P2 3.3.2. Analysis with modied values of Su and OCR (Class-C2)
(5.0 m from the ground surface). The Class-A prediction showed a It is well known that for clayey soils there is an effect of strain
slower dissipation rate, but in contrast the Class-C1 prediction rate on the undrained shear strength, Su (e.g., [2,3]). In addition, the
indicated much faster dissipation. The reasons for this difference soil at the site was micro-structured and has exhibited some
are considered to be: (1) the initial values of the hydraulic conduc- strain-softening behaviour, as indicated by the effective stress
tivities for the layers at depths between 1.5 m and 12.0 m shown in paths in Fig. 4. In the Class-A and Class-C1 analyses, the unconned
Table 2 are larger than the corresponding values listed in Table 1; compression strengths, qu, obtained from the undisturbed soil
and (2) the value of M was increased from 1.2 to 1.6. Due to these samples loaded at an axial strain rate of 1%/min, were used directly
changes, the predicted effective stress paths of the soils under the to calibrate the yield stress. Therefore, a possibility exists that the
embankment were altered. Fig. 14 shows an example comparison deduced values of Su, and hence the corresponding values of OCR,
of the effective stress paths for a soil element near point P2. might not be fully representative of eld conditions and it is possi-
124 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126

40
0 0.5 1 1.5 M = 1.6
0
M =1.2

Deviator stress, q (kPa)

30
Target q f
2

4
Depth, D (m)

20
Effective
6 stress
path
10 Initial
8 point
py py
(M = 1.6) (M = 1.2)
10 0
Test Site 0 10 20 30 40 50 60
Nearby Area
12
Effective mean stress, p' (kPa)

Fig. 13. Illustration of the effect of M on the size of the yield locus.
Fig. 12. Comparison of k values.

ble that they have been over-estimated. For this reason further
analyses with modied values of Su and OCR were conducted. This
50
further modication was made by multiplying the values of Su (Su-
A&C1 in Fig. 7) by a factor a, which was assigned values of 0.75, Stay here for
more than 100
0.8 and 0.85. The corresponding values of OCR were also re-esti- 40 days M = 1.2

Deviator stress, q (kPa)

mated. By comparing the various predictions with the eld mea-
sured values, it was found that a value of a = 0.85 results in the
best predictions, and this particular analysis is designated here as 30
M = 1.6
Class-C2. The values of Su and OCR for the Class-C2 analysis are
plotted in Figs. 7 and 8, where they are labelled Su-C2 and 20
OCR-C2, respectively. The factor a = 0.85 was applied to the mea-
sured values of Su and the results are designated as Corrected (S0u )
in Fig. 7. 10 Class-A
Class-C1
The results of the Class-C2 analysis are compared with the mea-
Class-C2
sured and other simulated results in Figs. 911 for settlements, lat-
0
eral displacements and excess pore water pressures, respectively. 0 10 20 30 40 50
From Fig. 9, it can be seen that the Class-C2 analysis resulted in a Effective mean stress, p' (kPa)
good simulation of the settlementtime curves, although it still
Fig. 14. Effective stress paths for a point near P2.
slightly under-estimated the settlement magnitudes. The results
in Fig. 10 indicate that the Class-C2 prediction slightly over-esti-
mated the lateral displacements, but overall the predictions are
As for the strain rate effect, Bjerrum [2] proposed a factor (l) to
very good. As for the excess pore water pressures, generally the re-
correct the eld vane shear strength (Su) used in slope stability
sults from the Class-C2 predictions are excellent, except for the
analysis. He correlated this factor with the plasticity index (PI) of
values at location P1 where the simulated dissipation rate was fas-
the soil. Tanaka [27] reported a comparison of eld vane shear
ter than the measured rate (Fig. 11a).
strength (Su) with values determined from laboratory unconned
compression test (qu/2) for seven different clayey soils in Japan
and indicated that for values of PI of about 4080, the average ratio
3.4. Signicance of the factor a Su/(qu/2) is about 1.0. For the soft Ariake clay samples from BH-6,
the PI values were 5276 (in the Nearby Area the PI values were
The main reasons for operative eld values of Su being less than 3270). With a PI of about 50, the corresponding Bjerrum correc-
the laboratory measured values are: (a) strain rate effects, and (b) tion factor (l) will be about 0.8, which is slightly smaller but com-
de-structuring or strain softening of the micro-structured (or sen- parable with the back-estimated value of a = 0.85. Using a value of
sitive) soil. Therefore a can be regarded as a correction factor for a of 0.8, the predictions are good, but not quite as good as those
the effects of strain rate and/or strain-softening. corresponding to a = 0.85.

Table 2
Redened model parameters based on results obtained for samples from BH-6.

Depth (m) Soil strata SPT N E (kPa) m j k M e0 ct (kN/m3) kv kh

(104 m/day)
0.01.5 Surface soil 0.15 0.025 0.25 1.6 1.50 16.0 6.0 9.1
1.54.0 Soft silty clay 0.15 0.107 1.07 1.6 3.71 13.1 10.4 15.6
4.06.0 0.15 0.119 1.19 1.6 2.88 13.9 17.3 25.9
6.08.0 0.15 0.084 0.84 1.6 2.72 14.1 16.5 24.7
8.09.8 0.15 0.066 0.66 1.6 1.91 15.7 7.0 10.6
Depth (m) Soil strata SPT N E (kPa) m / () c (kPa) M e0 ct (kN/m3) kv kh
9.812.0 12.015.0 Clayey sand 3 7500 0.15 35 20 0.80 18.0 2500 2500
12.015.0 Clayey sand 8 20,000 0.15 35 20 0.80 18.0 2500 2500
15.020.0 Clayey sand 15 37,500 0.15 35 20 0.70 19.0 2500 2500
 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126
When Bjerrums correction factor, l, was rst proposed, the ef-
fect of de-structuring or strain-softening of the soils considered
was not explicitly mentioned. However, since the values of l were
obtained by comparing the eld vane shear strengths with the
back-estimated eld mobilised strengths, and because it is gener-
ally recognised that all natural clayey soils are micro-structured
(e.g., [24]), it is highly likely that both the effects of strain rate
and strain-softening are included implicitly in the values of l pro-
posed by Bjerrum. Although further research is needed to establish
a more rational way to determine the strength reduction factor, a,
at present it is considered that Bjerrums strain rate correction fac-
tor can be used as a reasonable rst approximation of the correc-
tion factor for the values of Su.
The use of the Bjerrum correction has meant that the OCR val-
ues adopted for the Class-C2 analyses are lower than the values
determined from laboratory consolidation test results. It is noted
that consolidation test results only provide a single point on the
yield surface, and the initial yield function (or yield surface) has
to be extrapolated from this point by applying the appropriate
MCC theory. Hence there is a possibility that the use of the MCC
model together with a value of OCR determined from oedometer
testing may result in over-estimation of the eld value of Su. It is
emphasised again that calibrating the model, specically the pre-
dicted values of Su, using the measured values of Su corrected for
the strain rate effect (and possibly any softening arising from mi-
cro-structure in the clay), is an important step in making successful
predictions of the effects of embankment loading.
4. Discussion
The case study presented above has revealed at least three
important issues regarding prediction of the performance of earth
structures on soft soils.
(1) Adequate soil investigation: Of course detailed and appropri-
ate site investigation, soil testing and interpretation of the
test results are important in all geotechnical designs. This
case study has further emphasised this point. For any natural
deposit there may be localised variation of the soil proper-
ties, and there is additional risk if the design parameters
are estimated using only information obtained from nearby
areas rather than the specic site in question.
(2) Strength parameters of clayey soils: For most practical cases,
the effective stress strength parameters, e.g., the cohesion
(c0 ) and internal friction angle (/0 ) of a clayey soil, are usually
interpreted based on the Mohr-Coulomb failure criterion,
and normally nite values will be obtained for both c0 and
/0 . In cases where a no-cohesion soil model, such as MCC,
is used to predict the soil response, there is a possibility that
the strength parameter M may be under-estimated if the
value of /0 interpreted from the Mohr-Coulomb criterion is
adopted directly. This case study has demonstrated the
importance of determining the value of M from laboratory
tests following the appropriate effective stress path.
(3) Performance of the adopted constitutive model: To predict the
mechanical behaviour of a clayey deposit under embank-
ment loading using the nite element approach, it is essen-
tial to check whether the constitutive model, as well as the
adopted values of the model parameters, are able represent
well the strength and deformation characteristics of the sub-
soil. However, there is no well-established procedure or
method to conduct this kind of check. The results of this
study indicate that comparing the simulated prole of
undrained shear strength (Su) (under triaxial compression)
with measured values of Su may be a useful way to check
125
the suitability of a particular model and the values selected
for the model parameters. In this process, the effect of strain
rate and/or strain-softening on the values of Su has to be
properly considered. The logic behind this approach is that
provided parameters like the compression index and the
coefcient of consolidation are chosen correctly, the magni-
tudes of the predicted soil deformations are strongly inu-
enced by the yield stress (i.e., the size of the yield locus),
and for most constitutive models for clay, the predicted
value of Su is closely related to the adopted value of the ini-
tial yield stress.
It is proposed that a strength correction factor (a) should be ap-
plied to laboratory and eld measures of Su. Although further re-
search is needed to determine a values in a fundamental way,
the results of this study show that Bjerrums [2] empirical strain
rate correction factor (l) can be used as a rst approximation of
the value of a.
Although the MCC model adopted in this study does not con-
sider the anisotropic and time dependent behaviour possessed of
most natural clayey deposits, it has still resulted in quite good sim-
ulation (Class-C2) of the eld measured response. It is considered
that the corrected Su value most likely represents to sufcient
accuracy the average eld-mobilised value of Su under an embank-
ment load, and using this value of Su with an isotropic soil model
captures the most important features of the mechanical behaviour
of soft clayey soil, and generally results in acceptable predictions of
that behaviour. However, if the available site investigation data are
sufcient to reliably dene the necessary parameters for more
sophisticated soil models, such as those capturing anisotropy, in
principle use of these more sophisticated models should result in
improved predictions of soil behaviour.
5. Conclusions
Field measurements and numerical predictions of the behaviour
of a test embankment constructed on a deposit of soft Ariake clay
in Saga, Japan, have been presented. The numerical predictions
were conducted by nite element analysis (FEA) in which the
mechanical behaviour of soft clayey soil layers was represented
by the Modied Cam Clay model. These involved two types of pre-
diction: Class-A, conducted before; and Class-C, conducted after
the embankment was constructed. Comparisons were made be-
tween the various predictions and the eld measurements in terms
of settlements, lateral displacements and excess pore water pres-
sures in the subsoil.
The results of these comparisons indicate that the Class-A pre-
diction resulted in a poor simulation of the eld performance,
mainly due to over-estimation of the yield stresses (i.e., the sizes
of the initial yield loci) of the subsoils and under-estimation of
compressibility, hydraulic conductivity and the slope (M) of the
critical state line. The following observations are suggested by
the results of this study.
(a) Detailed on-site soil investigation, correct testing and appro-
priate interpretation of the test results are essential issues
for predicting the behaviour of an earth structure on a soft
clayey deposit.
(b) When using a soil model developed in the framework of
Critical State Soil Mechanics, the value of parameter M
should be determined directly from test result with an
appropriate effective stress path.
(c) Calibrating the yield stress by comparing the simulated pro-
le of undrained shear strength (Su) under triaxial compres-
sion with the measured data provides an efcient check on
126 J. Chai et al. / Computers and Geotechnics 48 (2013) 117126
4 [9] Chai J-C, Miura N, Shen S-L. Performance of embankments with and without
3 dx.doi.org/10.1520/GTJ103848.
Void ratio, e
2 [12] Grammatikopoulou A, Zdravkovic L, Potts DM. The effect of the yield and
Depth (m)
1 1.5-2.3 m
4.6-5.4 m
7.9-8.7 m
9.0-9.8 m
0 35462.
1 10 100
Vertical consolidation pressure 'v (kPa)
Fig. A1. e- logr0 v ) curves.
the suitability of the constitutive model, as well as the
adopted values of the model parameters, provided that a
correction for the effect of strain rate or strain-softening
on Su value is properly considered. The results also indicate
that Bjerrums strain rate correction factor can be adopted
as a rst approximation of the correction factor applied to
eld or laboratory measured values of Su.
Appendix A
Oedometer test results of undisturbed samples from BH-6 are
presented in Fig. A1.
References
[1] Ariake Sea Coastal Road Development Ofce (ASCRDO), Saga Prefecture. Report
on eld measurements and analyses: improvement project for national road
no. 444, Saga, Japan; 2011 [in Japanese].
[2] Bjerrum L. Embankment on soft ground: SOA report. In: Proc specialty conf on
performance of earth and earth-supported structures, ASCE, vol. 2; 1972. p. 1
54.
[3] Bjerrum L. Embankment on soft ground: SOA report. In: Proc 8th conf soil
mech and found eng, vol. 3, Moscow, USSR; 1973. p. 11159.
[4] Brand EW. Predicted and observed performance of an embankment built to
failure on soft clay. Geotech Eng, Southeastern Asian Geotech Soc
1991;22(1):2342.
[5] Britto AM, Gunn MJ. Critical state soil mechanics via nite elements. Ellis
Horwood Limited; 1987. p. 486.
[6] Chai JC, Miura N, Bergado DT, Long PV. Finite element analysis of embankment
failure on soft subsoil. Geotech Eng, Southeast Asian Geotech Soc
1997;28(2):24976.
[7] Chai J-C. Interaction between grid reinforcement and cohesive-frictional soil
and performance of reinforced wall/embankment on soft ground. PhD
dissertation, Asian Institute of Technology, Bangkok, Thailand; 1992.
[8] Chai J-C, Miura N. Investigation on some factors affecting vertical drain
behavior. J Geotech Geoenviron Eng, ASCE 1999;125(3):21626.
reinforcement on soft subsoil. Can Geotech J 2002;39(8):83848.
[10] Chai J-C, Jia R, Hino T. Anisotropic consolidation behavior of Ariake clay from
three different CRS tests. Geotech Test J, ASTM 2012;35(6):19. http://
[11] Grammatikopoulou A, Zdravkovic L, Potts DM. General formulation of two
kinematic hardening constitutive models with a smooth elastoplastic
transition. Int J Geomech, ASCE 2006;6(5):291302.
plastic potential deviatoric surfaces on the failure height of an embankment.
Gotechnique 2007;57(10):795806.
[13] Igaya Y, Hino T, Chai J-C. Behavior of trial embankments on soft Ariake clay
deposits with and without column improvement. In: Proceedings
international symposium on geotechnical and geosymthetics engineering:
challenges and opportunities on climate change, Bangkok, Thailand; 2010. p.
1000 [14] Japanese Society of Civil Engineer (JSCE). Manual for structural design of
Japanese Railway, Tokyo; 1986.
[15] Karim MR, Gnanendran CT, Lo S-CR, Mak J. Predicting the long-term
performance of a wide embankment on soft soil using an elasticviscoplastic
model. Can Geotech J 2010;47(2):24457.
[16] Karstunen M, Yin Z-Y. Modelling time-dependent behaviour of Murro test
embankment. Gotechnique 2010;60(10):73549. http://dx.doi.org/10.1680/
geot.8.P.027-735.
[17] Lambe TW. Predictions in soil engineering. Gotechnique 1973;23:149202.
[18] Manivannan G, Karim MR, Gnanendran CT, Lo S-CR. Calculated and observed
long term performance of Leneghans embankment. Geomech Geoeng
2011;6(3):195207.
[19] Mayne PE, Kulhawy FH. K0-OCR relationships in soils. J Geotech Eng, ASCE
1982;108(GT6):85172.
[20] Oliveira PJV, Lemos LJL, Coelho PALF. Behavior of an atypical embankment on
soft soil: eld observations and numerical simulation. J Geotech Geoenviron
Eng 2010;136(1):3547.
[21] Oliveira PJV, Lemos LJL. Numerical predictions of the behaviour of soft clay
with two anisotropic elastoplastic models. Comput Geotech
2010;38(5):598611.
[22] Poulos HG, Lee CY, Small JC. Prediction of embankment performance on
Malaysian marine clays. In: Hudson RR, Toh CT, Chan SF, editors. Proceedings
of the international symposium on trial embankments on Malaysian marine
clays, Kuala Lumpur, Malaysia, vol. 2. Malaysia: The Malaysian Highway
Authority, Kuala Lumpur; 1989. p. 4/14/10.
[23] Roscoe KH, Burland JB. On the generalized stressstrain behavior of wet clay.
In: Heyman J, Leckie FA, editors. Engineering plasticity. Cambridge University
Press; 1986. p. 535609.
[24] Schmertmann JH. The mechanical aging of soils. J Geotech Eng, ASCE
1991;117(9):1288330.
[25] Shimoyama S, Matsuura H, Hino T. Topography, geology of the saga district.
Quadrangle series, 1:50,000, Geological Survey of Japan, AIST; 2010. p. 17
[chapter 1, in Japanese].
[26] Sloan SW. Substepping schemes for numerical integration of elato-plastic
stressstrain relations. Int J Numer Methods Eng 1987;24:893911.
[27] Tanaka H. Vane shear strength of a Japanese marine clay and applicability of
Bjerrum correction factor. Soils Found 1994;34(3):3948.
[28] Tavenas F, Tremblay M, Larouche G, Lerouei lS. In situ measurement of
permeability in soft clays. In: Proceedings ASCE special conference on use of
in-situ tests in geotechnical engineering, Blacksburg; 1986. p. 103448.
[29] Taylor DW. Fundamentals of soil mechanics. New York: John Wiley & Sons,
Inc.; 1948.
[30] Wroth CP. The predicted performance of soft clay under a trial embankment
loading based on the Cam-clay model. In: Gudehus G, editor. Finite elements in
geomechanics. Wiley; 1977. p. 191208.
[31] Yin J-H, Zhu J-G, Graham J. A new elastic viscoplastic model for time
dependent behaviour of normally and over consolidated clays: theory and
verication. Can Geotech J 2002;39(1):15773.